Research map / Label-efficient and robust learning / Recognition and long-tail
Uncertainty Modeling for Out-of-Distribution Generalization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace deterministic feature statistics with sampled Gaussian statistics using batch-estimated, channel-specific uncertainty | ResNet18 · PACS | Average accuracy (%): 79.0 -> 84.1 | Helped |
| Replace deterministic feature statistics with sampled Gaussian statistics using batch-estimated, channel-specific uncertainty | DeepLab-v2 with ResNet101 · GTA5 to Cityscapes | mIOU (%): 37.0 -> 43.1 | Helped |
| Replace deterministic feature statistics with sampled Gaussian statistics using batch-estimated, channel-specific uncertainty | ResNet50 · Market to Duke | mAP (%): 25.8 -> 32.0 | Helped |
| Replace deterministic feature statistics with sampled Gaussian statistics using batch-estimated, channel-specific uncertainty | ResNet50 · Duke to Market | Rank-1 (%): 54.7 -> 63.7 | Helped |
| Add DSU during training | ResNet50 · ImageNet-C | mCE (%): 76.2 -> 73.4 | Helped |
| Add DSU to APR | APR · ImageNet-C | mCE (%): 65.0 -> 64.1 | Helped |
| Add DSU during training | ResNet18 · Office-Home | Average accuracy (%): 64.4 -> 66.1 | Helped |
| Insert DSU at positions 0-5 rather than leaving the model unaugmented (ablation) | ResNet · PACS | Average accuracy (%): 79.0 -> 84.1 | Helped |
| Use the adaptive Gaussian uncertainty scope instead of fixed Gaussian shifts (ablation) | DSU · PACS | Average accuracy (%): 76.9 -> 84.1 | Helped |
| Use the adaptive Gaussian uncertainty scope instead of uniform shifts (ablation) | DSU · PACS | Average accuracy (%): 81.9 -> 84.1 | Helped |
| Use channel-specific uncertainty rather than sharing one uncertainty distribution across channels (ablation) | DSU · PACS | Average accuracy (%): 80.2 -> 84.1 | Helped |
| Vary the training batch size while using DSU (ablation) | DSU · PACS | Average accuracy (%): 81.0 -> 84.9 (+3.9); 80.2 -> 84.5 (+4.3); 79.0 -> 84.1 (+5.1); 77.8 -> 82.1 (+4.3); 75.6 -> 80.4 (+4.8) | Mixed |
Closest papers by meaning
- Feature Contamination: Neural Networks Learn Uncorrelated Features and Fail to Generalize (ICML 2024 · similarity 0.88)
- Deep Stable Learning for Out-of-Distribution Generalization (CVPR 2021 · similarity 0.88)
- Assaying Out-Of-Distribution Generalization in Transfer Learning (NeurIPS 2022 · similarity 0.88)
- Uncertainty-Guided Model Generalization to Unseen Domains (CVPR 2021 · similarity 0.88)
- Can Biases in ImageNet Models Explain Generalization? (CVPR 2024 · similarity 0.86)
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.